The present invention is related to audio processing and in particular to a concept for decoding an encoded audio signal using reduced computational resources.
The ‘Unified speech and audio coding” (USAC) standard [1], standardizes a harmonic bandwidth extension tool, HBE, employing a harmonic transposer, and which is an extension of the spectral band replication (SBR) system, standardized in [1] and [2], respectively.
SBR synthesizes high frequency content of bandwidth limited audio signals by using the given low frequency part together with given side information. The SBR tool is described in [2], enhanced SBR, eSBR, is described in [1]. The harmonic bandwidth extension HBE which employs phase vocoders is part of eSBR and has been developed to avoid the auditory roughness which is often observed in signals subjected to copy-up patching, as it is carried out in the regular SBR processing. The main scope of HBE is to preserve harmonic structures in the synthesized high frequency region of the given audio signal while applying eSBR.
Whereas an encoder can select the usage of the HBE tool, a decoder which is conform to [1] shall provide decoding and applying HBE related data.
Listening tests [3] have shown that using HBE will improve perceptual audio quality of decoded bitstreams according to [1].
The HBE tool replaces the simple copy-up patching of the legacy SBR system by advanced signal processing routines. These necessitate a considerable amount of processing power and memory for filter states and delay lines. On the contrary the complexity of the copy-up patching is negligible.
The observed complexity increase with HBE is not a problem for personal computer devices. However, chip manufactures designing decoder chips are demanding rigid and low complexity constraints regarding computational workload and memory consumption. Otherwise, HBE processing is desired in order to avoid auditory roughness.
USAC-bitstreams are decoded as described in [1]. This implies necessarily the implementation of a HBE decoder tool, as described in [1], 7.5.3. The tool can be signaled in all codec operating points which contain eSBR processing. For decoder devices which fulfill profile and conformance criteria of [1] this means that the overall worst case of computational workload and memory consumption increases significantly.
The actual increase in computational complexity is implementation and platform dependent. The increase in memory consumption per audio channel is, in the current memory optimized implementation, at least 15 kWords for the actual HBE processing.